80% enterprise AI leaders are expecting AI to span across businesses: AI governance, risk & compliance [GRC] and lifecycle management practices are becoming critical

In a recent HFS study, more than 80% of enterprise AI leaders said they were expecting artificial intelligence (AI) and machine language (ML) to span all aspects of their businesses within the next two to five years. Out of this group, 52% are planning to invest in AI and ML across their businesses in the current two-year period. When AI becomes pervasive at that scale of expansion, governance and management of AI solutions’ lifecycles become a key challenge for all enterprise AI leaders who have embarked on this journey. The dangers of these autonomous decision-and-action models losing recency and relevance can be enormous, especially when these solutions are put in production in critical and end-to-end business contexts. The other aspect is the increasing enforcement of standards and regulations of government bodies on the data privacy and fairness of AI use cases; for example, the ISO/ IEC JTC-1, IEEE P7000 series of standards, and FAIR AI guidelines. Standards will ensure transparency, consistency, predictability, and good governance; enforcing, auditing, monitoring, and maintaining them require dedicated efforts.

When the number of smart bots go up from 10’s to 1000’s, AI GRC as a service becomes industry’s need of the hour

As a senior client leader in a large high-tech company based in the APAC region explained, the problems of governance and maintenance become exponential as the deployment of AI use-cases starts scaling. It’s one thing to manage 5 to 10 bots in the POC level of a use case, for example. But you need to scale it as much as possible to get end-to-end outcomes and benefits from the solution. When the number of bots is in the thousands, GRC for them requires regular calibrations of accuracy and performance, orchestration management, and data quality checks for biases, to monitor and control health, behavior and efficacy of decisions and actions of the AI modules. It’s more than a multiplier problem; it’s an exponential one.

Therefore, for AI, maintenance is much more than post-production tech support. In fact, most AI solutions can be quite self-managed if the deployment is done right and most scenarios and exceptions are covered in the use-cases, making the tech maintenance piece the least of clients’ problems. But, very few of these solutions are autonomously “self-learning.” Self-learning AI can autonomously adjust weights of different nodes that represent the resolution knowledge items in the knowledge graph, based on the efficacy of usages and feedback loops—it literally learns and augments itself “on the job.” Such self-augmenting AI platforms are still more exception than mainstream. So, the patterns and models the inference engines use in production deployments may not always reflect the most recent reality. To maintain relevance and recency, maintenance teams will have to watch out for any potential degradation in accuracy or coverage of scenarios. Whenever such triggers show up, they have to retrain the model with all the new and relevant datasets and knowledge artifacts.

Act now: deploy strong AI GRC as a service

Enterprise AI leaders must plan for AI solutions’ governance, risk, and compliance (GRC) requirements so that they can continue to gain sustainable value from the expensive AI and automation projects that usually have payback periods of at least three to five years. Here are three actions you must take up.

Exhibit 1: Action items for next Monday morning

The Bottom Line: Enterprise AI leaders must plan to use autonomous AI-based AI governance and lifecycle management as a service.

Enterprise AI leaders must plan to use AI GRC and lifecycle management offered as a service as an integral part of AI scaling and adoption plans.  That way, they can stay ahead of the curve, in a compliant and risk-mitigated manner, while leveraging innovations and net new practices, in the ever-evolving and expanding realms of AI adoption.